EE380 Spring 2004 Sample Final Exam

This sample exam is derived from the nal in Spring 2002. Note that it is only a sample; although general topic coverage will be similar, the precise material that you will be questioned on will be somewhat different. 1. We discussed several different types of network topologies for parallel processing. Which of the following statements about networks is false? Direct Connection (aka, fully connected) networks offer very good latency and bandwidth, but do not scale well Channel Bonding refers to the concept of using multiple lower-speed networks in parallel to achieve the performance of a more-expensive higher-speed network Flat Neighborhood Networks (FNNs) make use of the fact that a PC can have more than one NIC For communication-intensive parallel computing, networks built using switches generally outperform those built using hubs


2.

Most applications that run well on a cluster also will run well on a grid or farm

Which of the following four statements about the memory hierarchy is false? L1 cache generally is faster to access than L2 cache Temporal Locality refers to the concept of objects with addresses near each other being referenced over a shor t period of time Larger cache line sizes take better advantage of Spatial Locality A high cache hit ratio usually is a good thing None of the above four statements is false

3.

Which of the following I/O mechanisms requires the least hardware support? Polled Interrupt-driven DMA

4.

Why might one use a larger line size in an instruction cache than in a data cache?

Larger line sizes are better for spatial locality and code tends to have much higher spatial locality than data.

Page 1 of 7

EE380 Spring 2004 Sample Final Exam 5. Consider the two MIPS subset implementations shown at the back of this test. Which of the following four statements about how pipelining changes the architecture is false? The ALU used to add 4 to the PC could be the same circuit in both implementations The ALU used for operations like And and Slt could be the same circuit in both implementations The Data Memory (data cache) module could be the same circuit in both implementations The Instruction Memory (instruction cache) module could be the same circuit in both implementations

None of the above four statements is false; in fact, all of the modules can be the same circuits in both implementations because pipelining only adds buffers, changes/adds some datapaths, and modies the control logic

6.

Consider the pipelined MIPS subset implementation shown at the back of this test. Given this design, briey explain the performance problem that is caused by the Beq instruction.

The branch in branch-if-equal might, or might not, be taken; thus, it causes a control hazard in that we dont know which instruction should enter the pipeline immediately after Beq. Further, if the branch is taken, the target address might not have been computed quickly enough to avoid inserting a pipeline bubble.

7.

In the supercomputing eld, there is an old joke that the highest bandwidth communication medium is a station wagon full of magnetic tapes. Modernizing the joke, a typical minivan could hold about 10,000 tape cartridges, each of which stores about 10GB of data. Assume that the total time taken to write the tapes, drive to the other computer, and read the tapes is about 100,000 seconds (just under 28 hours). Which of the following more conventional networks is closest to the same bandwidth? (Hint: b/s isnt the same as B/s.) 9600 baud modem (phone line connection) 10 Mb/s Ethernet 100 Mb/s Fast Ethernet 1 Gb/s Gigabit Ethernet

10 Gb/s Ethernet (Note b is bits, B is Bytes)

Page 2 of 7

EE380 Spring 2004 Sample Final Exam 8. Consider executing each of the following code sequences on the pipelined MIPS implementation given at the end of this test (do not assume any datapaths not shown in the diagram). Incidentally, both code sequences produce the same nal results. Which of the following statements best describes the execution times you would expect to observe? (A) Add $t0,$t1,$t2 Add $t5,$t6,$t7 Add $t3,$t0,$t4 Add $t0,$t1,$t2 Add $t3,$t0,$t4 Add $t5,$t6,$t7

(B)

(A) would be faster than (B) (B) would be faster than (A) (A) would take the same number of clock cycles as (B)

9.

Consider executing each of the following code sequences on the pipelined MIPS implementation given at the end of this test. Also consider executing them on this design with value forwarding logic and datapaths added. Which of the following statements best describes how the forwarding logic would alter the execution times? (A) Add $t0,$t1,$t2 Add $t0,$t0,$t3 Add $t0,$t0,$t4 Add $t0,$t1,$t2 Add $t3,$t4,$t5 Add $t6,$t7,$t8

(B)

Neither (A) nor (B) is affected by forwarding (A) is not affected, (B) would be faster using forwarding

(A) would be faster using forwarding, (B) is not affected Both (A) and (B) would be faster using forwarding The execution time improvements due to forwarding depend on the values in the registers, not on the instructions being executed; thus, it is impossible to say how execution times for (A) and (B) are affected

Page 3 of 7

EE380 Spring 2004 Sample Final Exam 10. What is a pipeline bubble? Precisely what is in a pipeline stage that is executing a bubble?

A pipeline bubble is one or more pipeline stages that do not contain useful instructions; i.e., they contain NOPs or other side-effect-free instructions. Bubbles typically are introduced to accomodate various types of pipeline hazzards.

11.

The following diagram shows the internals of the AMD K6-2 processor. According to the diagram, which of the following techniques is not used in this design?

Separate translation lookaside buffers for code and data Separate L1 caches for code and data

Separate L2 caches for code and data Superscalar execution of two SWAR (MMX or 3DNow!) instructions per clock cycle Superscalar execution of integer operations

Page 4 of 7

EE380 Spring 2004 Sample Final Exam 12. Suppose that x is a value represented in 8-bit 2s complement binary notation. Oddly enough, the result of computing -x cannot be represented in 8-bit 2s complement binary notation. What is the decimal value of x?

2s complement numbers go one value more negative than positive, from -128..+127 for 8 bits. Since there is no -(-128), x must be -128.

13.

You have two different ways to implement a particular system. In one version, the cache is on the processor chip; in the other the cache is slower but much larger because it is off chip. Suppose that the rst design achieves 1 cycle access latency with a hit rate of 80% and the second design achieves 2 cycle latency with a hit rate of 99%. In either case, main memory has a latency of 10 cycles. Which design is faster and by how much? vs. 2*99+10*1=208; the second

This is really 1*80+10*20=280 design is faster.

14.

Which of the following four statements about control logic in general is false? Outputs are enabled at the end of each clock cycle Inputs are latched at the end of each clock cycle The choice of how instructions are encoded as bit patterns can change the complexity of the control logic Although random logic implementations of control may result in faster clock rates than schemes using gate arrays, memory, or microcode, it is more difcult to make changes to x bugs in random logic implementations None of the above four statements is false

Page 5 of 7

EE380 Spring 2004 Sample Final Exam 15. In terms of A, B, and C, what does the gate shown below do?
B

A 0 0 1 1

B 0 1 0 1

C Z 0 Z 1

16.

What does it mean when we say a cache is two-way associative?

It means that the cache uses set associativity in which each line of data from memory can go into either of two line positions within the cache.

17.

Consider the single-cycle implementation at the end of this test. Which of the following statements about how we said MIPS instructions could be implemented is false? Although the ALU operation control lines would be different, the Or and Sub instructions are otherwise handled identically The Sw instruction would set ALUSrc to select the lower input to the Mux

The Shift function unit is used by the Slt instruction The result of the Add ALU in the upper right of the diagram is meaningless garbage for an Add instruction All the above four statements are true.

Page 6 of 7

EE380 Spring 2004 Sample Final Exam 18. Consider the single-cycle implementation at the end of this test. Which of the following control lines is a dont care while executing a Beq instruction? PCSrc ALUSrc RegWrite

MemtoReg ALU operation

19.

Which of the following typically is not found inside the case of an Intel Pentium 4 or AMD Athlon PC? PCI bus DRAM CPU

Interociter One or more fans

20. Jump

What is an instruction that loads an immediate value into the PC generally called?

Page 7 of 7